Analytically Tractable Climate–Carbon Cycle Feedbacks Under 21St Century Anthropogenic Forcing

Analytically tractable climate-carbon cycle feedbacks under 21st century anthropogenic forcing

Lade, Steven J.; Donges, Jonathan F.; Fetzer, Ingo; Anderies, John M.; Beer, Christian; Cornell, Sarah E.; Gasser, Thomas; Norberg, Jon; Richardson, Katherine; Rockström, Johan; Steffen, Will

Published in: Earth System Dynamics

DOI: 10.5194/esd-9-507-2018

Publication date: 2018

Document version Publisher's PDF, also known as Version of record

Document license: CC BY

Citation for published version (APA): Lade, S. J., Donges, J. F., Fetzer, I., Anderies, J. M., Beer, C., Cornell, S. E., Gasser, T., Norberg, J., Richardson, K., Rockström, J., & Steffen, W. (2018). Analytically tractable climate-carbon cycle feedbacks under 21st century anthropogenic forcing. Earth System Dynamics, 9(2), 507-523. https://doi.org/10.5194/esd-9-507- 2018

Download date: 01. Oct. 2021 Earth Syst. Dynam., 9, 507–523, 2018 https://doi.org/10.5194/esd-9-507-2018 © Author(s) 2018. This work is distributed under the Creative Commons Attribution 4.0 License.

Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

Steven J. Lade1,2,3, Jonathan F. Donges1,4, Ingo Fetzer1,3, John M. Anderies5, Christian Beer3,6, Sarah E. Cornell1, Thomas Gasser7, Jon Norberg1, Katherine Richardson8, Johan Rockström1, and Will Steffen1,2 1Stockholm Resilience Centre, Stockholm University, Stockholm, Sweden 2Fenner School of Environment and Society, The Australian National University, Australian Capital Territory, Canberra, Australia 3Bolin Centre for Climate Research, Stockholm University, Stockholm, Sweden 4Potsdam Institute for Climate Impact Research, Potsdam, Germany 5School of Sustainability and School of Human Evolution and Social Change, Arizona State University, Tempe, Arizona, USA 6Department of Environmental Science and Analytical Chemistry (ACES), Stockholm University, Stockholm, Sweden 7International Institute for Applied Systems Analysis, Laxenburg, Austria 8Center for Macroecology, Evolution, and Climate, University of Copenhagen, Natural History Museum of Denmark, Copenhagen, Denmark Correspondence: Steven J. Lade ([email protected])

Received: 1 September 2017 – Discussion started: 12 September 2017 Revised: 23 April 2018 – Accepted: 27 April 2018 – Published: 17 May 2018

Abstract. Changes to climate–carbon cycle feedbacks may signiﬁcantly affect the Earth system’s response to greenhouse gas emissions. These feedbacks are usually analysed from numerical output of complex and arguably opaque Earth system models. Here, we construct a stylised global climate–carbon cycle model, test its output against comprehensive Earth system models, and investigate the strengths of its climate–carbon cycle feedbacks analytically. The analytical expressions we obtain aid understanding of carbon cycle feedbacks and the operation of the carbon cycle. Speciﬁc results include that different feedback formalisms measure fundamentally the same climate–carbon cycle processes; temperature dependence of the solubility pump, biological pump, and CO2 solubility all contribute approximately equally to the ocean climate–carbon feedback; and concentration–carbon feedbacks may be more sensitive to future climate change than climate–carbon feedbacks. Simple models such as that developed here also provide “workbenches” for simple but mechanistically based explorations of Earth system processes, such as interactions and feedbacks between the planetary boundaries, that are currently too uncertain to be included in comprehensive Earth system models.

1 Introduction pogenic carbon emissions each year (Le Quéré et al., 2016). These feedbacks are expected to weaken in the future, am- The exchanges of carbon between the atmosphere and other plifying the effect of anthropogenic carbon emissions on cli- components of the Earth system, collectively known as the mate change (Ciais et al., 2013). The degree to which they carbon cycle, currently constitute important negative (damp- will weaken, however, is highly uncertain, with Earth sys- ening) feedbacks on the effect of anthropogenic carbon emis- tem models predicting a wide range of land and ocean car- sions on climate change. Carbon sinks in the land and the ocean each currently take up about one-quarter of anthro-

Published by Copernicus Publications on behalf of the European Geosciences Union. 508 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

Climate change Temperature, T

+ Radiative -

- forcing Land carbon Ocean carbon Total vegetation + + Ocean mixed layer, cm and soils, ct (Total ocean, cM) Atmosphere

carbon, ca - -

Figure 1. Climate–carbon cycle feedbacks and state variables as represented in the stylised model introduced in this paper. Carbon stored on land in vegetation and soils is aggregated into a single stock ct. Ocean mixed layer carbon, cm, is the only explicitly modelled ocean stock of carbon; though to estimate carbon cycle feedbacks we also calculate total ocean carbon (Eq.7). bon uptakes even under identical atmospheric concentration upper layer of the ocean to then be transported deeper by or emission scenarios (Joos et al., 2013). physical and biological processes. The concentration–carbon Here, we develop a stylised model of the global carbon feedbacks are generally negative, dampening the effects of cycle and its role in the climate system to explore the poten- anthropogenic emissions. (iii) In the land climate–carbon tial weakening of carbon cycle feedbacks on policy-relevant feedback, higher temperatures, along with other associated timescales (< 100 years) up to the year 2100. Whereas com- changes in climate, generally lead to decreased storage on prehensive Earth system models are generally used for pro- land at the global scale, for example due to the increase in jections of climate, models of the Earth system of low com- respiration rates with temperature. (iv) In the ocean climate– plexity are useful for improving mechanistic understanding carbon feedback, higher temperatures generally lead to re- of Earth system processes and for enabling learning (Randers duced carbon uptake by the ocean, for example due to de- et al., 2016; Raupach, 2013). Compared to comprehensive creasing solubility of CO2. The climate–carbon feedbacks Earth system models, our model has far fewer parameters, are generally positive, amplifying the effects of carbon emis- can be computed much more rapidly, can be more rapidly sions. understood by both researchers and policymakers, and is We begin by introducing our stylised carbon cycle model even sufficiently simple that analytical results about feed- and testing its output against historical observations and fu- back strengths can be derived. Compared to previous stylised ture projections of Earth system models. Having thus estab- models (Gregory et al., 2009; Joos et al., 1996; Meinshausen lished the model’s performance, we introduce different for- et al., 2011a,c; Gasser et al., 2017a), our model features sim- malisms used to quantify climate–carbon cycle feedbacks ple mechanistic representations, as opposed to parametric fits and describe how they can be computed both numerically to Earth system model output, of aggregated carbon uptake and analytically from the model. We use our results to analyt- both on land and in the ocean. Our stylised and mechanisti- ically study the relative strengths of different climate–carbon cally based climate–carbon cycle model also offers a work- cycle feedbacks and how they may change in the future, as bench for investigating the influence of mechanisms that are well as to compare different feedback formalisms. We con- at present too uncertain, poorly defined, or computationally clude by speculating on how this stylised model could be intensive to include in current Earth system models. Such used as a “workbench” for studying a range of complex Earth stylised models are valuable for exploring the uncertain but system processes, especially those related to the biosphere. potentially highly impactful Earth system dynamics such as interactions between climatic and social tipping elements 2 Model formulation (Lenton et al., 2008; Kriegler et al., 2009; Schellnhuber et al., 2016) and the planetary boundaries (Rockström et al., 2009; There is well-developed literature on stylised models used Steffen et al., 2015). for gaining a deeper understanding of Earth system dynamics Analyses of climate–carbon cycle feedbacks convention- and even for successfully emulating the outputs of compre- ally distinguish four different feedbacks (Fig.1)(Friedling- hensive coupled atmosphere–ocean and carbon cycle mod- stein, 2015; Ciais et al., 2013). (i) In the land concentration– els (Anderies et al., 2013; Gregory et al., 2009; Joos et al., carbon feedback, higher atmospheric carbon concentration 1996; Meinshausen et al., 2011a,c; Gasser et al., 2017a). We generally leads to increased carbon uptake due to the fertili- developed a combination of existing models and new formu- sation effect, in which increased CO2 stimulates primary pro- lations to construct a global climate–carbon cycle model with ductivity. (ii) In the ocean concentration–carbon feedback, the following characteristics. physical, chemical, and biological processes interact to sink carbon. Atmospheric CO2 dissolves and dissociates in the 1. The model includes processes relevant to the carbon cycle and its interaction with climate on the policy-

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 509

relevant timescale of the present to the year 2100. climate-change-related effects on global NPP occurring si- Stylised carbon cycle models often do not, for exam- multaneously with carbon dioxide changes, for example, pre- ple, include explicit representations of the solubility or cipitation and temperature effects, in addition to fertilisation biological pumps. effects. The curvature of the log function represents limitations to NPP such as changing carbon-use efficiency (Körner, 2. The model produces quantitatively plausible output for 2003) or nutrient limitations (Zaehle et al., 2010). Constant carbon stocks and temperature changes. climate sensitivity is also a key assumption, otherwise the relative weight of climate and CO2 effects on NPP would 3. All parameters have a direct biophysical or biogeo- change. chemical interpretation, although these parameters may At the same time, carbon loss from the world’s soils be at an aggregated scale (for example, a parameter for through respiration, R, is expected to increase at higher the net global fertilisation effect, rather than leaf phys- global mean surface temperature, 1T . We approximate the iological parameters). We avoid models or model com- net temperature response of global soil respiration using the ponents constructed by purely parametric fits, such as 1T /10 Q formalism R(1T ) = R Q c /c (Xu and Shang, impulse response functions, to historical data or projec- 10 0 R t t0 2016), where Q is the proportional increase in respira- tions of Earth system models (Kamiuto, 1994; Gasser R tion for a 10 K temperature increase. We assume that pre- et al., 2017b; Joos et al., 1996; Harman et al., 2011; Gre- industrial soil respiration is balanced by pre-industrial NPP, gory et al., 2009; Meinshausen et al., 2011a). R0 = NPP0. To avoid introducing multiple pools of carbon 4. The model is sufficiently simple that calculation of into the model, we also have to assume that global soil respi- the model’s feedback strengths is readily analytically ration is proportional to total land carbon (rather than soil tractable. This tractability may come at the expense carbon). Respiration in our model also implicitly includes of complexity, for example multiple terrestrial car- other carbon-emitting processes such as wildfires or insect bon compartments, or accuracy at millennial or longer disturbances. timescales (Lenton, 2000; Randers et al., 2016). It follows that the change in global terrestrial carbon storage is Building on the work of Anderies et al.(2013), we con- dc c NPP structed a simple model with globally aggregated stocks of t = NPP 1 + K log a − 0 Q1T /10c − LUC(t). dt 0 C c c R t atmospheric carbon in the form of carbon dioxide, ca; terres- a0 t0 trial carbon, including vegetation and soil carbon, c ; and dis- t In this expression we have also included loss of terrestrial solved inorganic carbon (DIC) in the ocean mixed layer, c . m carbon due to land use emissions LUC(t). We rearrange this The model’s fourth state variable is global mean surface tem- expression to give perature relative to pre-industrial, 1T = T − T0. Compared to Anderies et al.(2013), our model includes more realistic dc NPP t = 0 1T /10 − − representation of terrestrial and ocean processes but without QR [K(ca,1T ) ct] LUC(t), (2) dt ct0 increase in model complexity, as well as time lags for climate response to CO2. where the terrestrial carbon carrying capacity is We now describe the dynamics of the land carbon stock, ca 1 + KC log the ocean carbon stock, and atmospheric carbon and temper- K(c ,1T ) = ca0 c . (3) ature in our model. a 1T /10 t0 QR For model simplicity, we do not explicitly model fac- 2.1 Land tors affecting terrestrial carbon uptake such as seasonality, Net primary production (NPP) is the net uptake of carbon species interactions, species functionality, migration, and re- from the atmosphere by plants through photosynthesis. NPP gional variability. is expected to increase with concentration of atmospheric carbon dioxide ca. A simple parameterisation of this so- 2.2 Ocean called fertilisation effect is “Keeling’s formula” for global NPP (Bacastow et al., 1973; Alexandrov et al., 2003): In the upper-ocean mixed layer, mixing processes allow exchange of carbon dioxide with the atmosphere. The solubility ca and biological pumps then transport carbon from the mixed NPP(ca) = NPP0 1 + KC log . (1) ca0 layer into the deep ocean. Since the residence time of deep ocean carbon is several centuries, we explicitly only model Throughout this article, the subscript “0” denotes the value the dynamics of upper-ocean carbon while the deep ocean of the quantity at a pre-industrial equilibrium, and “log” de- is treated merely as an extremely large carbon reservoir. We notes natural logarithm. Keeling’s formula incorporates all include the effects of ocean carbon chemistry, the solubility www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 510 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing and biological pumps, and ocean–atmosphere diffusion on back to the mixed layer by ocean circulation. We neglect upper-ocean mixed layer carbon. deposition of organic carbon to the sea floor and the long Ocean uptake of carbon dioxide from the atmosphere is timescale variations in the biological pump that may have − chemically buffered by other species of DIC such as HCO3 contributed to glacial–interglacial cycles (Sigman and Boyle, 2− 2000). We therefore add an additional term B(1T )−B(0) to and CO3 , which are produced when dissolved CO2 reacts with water. The reaction of CO2 with water, producing these the transport of carbon from the ocean mixed layer to the other species, reduces the partial pressure of CO2 in water al- deep ocean. Organic carbon that does not sink to the deep lowing for more ocean CO2 uptake before equilibrium with ocean is rapidly respired back to forms of inorganic carbon; the atmosphere is achieved. The Revelle factor, r, is defined the ocean mixed layer stock of organic carbon is therefore as the ratio of the proportional change in carbon dioxide con- small, around 3 PgC (Ciais et al., 2013), and we do not count tent to the proportional change in total DIC (Sabine et al., it in the model’s carbon balance. 2004; Goodwin et al., 2007). For simplicity, we assume a By combining the expressions for the solubility and bio- constant Revelle factor, except for the temperature depen- logical pumps with ocean–atmosphere carbon dioxide diffu- dence, DT , of the solubility of CO2 in seawater. Therefore sion, we obtain the rate of change of ocean mixed layer DIC, CO2 diffuses between the atmosphere and ocean mixed layer cm: at a rate proportional to dcm Dcm0 = (ca − p(cm,1T )) ca − p(cm,1T ), (4) dt rp(cm0,0) − − − − + where w0(1 wT 1T )(cm cm0) B(1T ) B(0). (6) r cm 1 The coefficient of the first term was chosen such that 1/D is p(cm,1T ) = ca0 , (5) cm0 1 − DT 1T the timescale on which carbon dioxide diffuses between the atmosphere and the ocean mixed layer (that is, the deriva- since at pre-industrial equilibrium p(cm0,0) = ca0. tive of the first term with respect to cm, evaluated at the pre- There are two main mechanisms by which carbon is trans- industrial equilibrium, is D). ported out of the upper-ocean mixed layer into the deep The carbon content of the deep ocean does not explicitly ocean stocks: the solubility and biological pumps. In the sol- enter Eq. (6). To evaluate ocean carbon feedbacks, however, ubility pump, overturning circulations exchange mixed layer we require the change in total ocean carbon content cM com- and deep ocean water. We assume that the large size of the pared to pre-industrial conditions. We calculate this as ocean deep ocean means its carbon concentrations are negligibly mixed layer carbon plus carbon transported to the deep ocean changed over the 100-year timescales relevant for the model. by the solubility and biological pumps: The net transport of carbon to the lower ocean by the solubility pump can therefore be represented by t Z h 1cM = 1cm + w0(1 − wT 1T )(cm(t) − cm0) w0(1 − wT1T )(cm − cm0), i where w0 is the (proportional) rate at which mixed layer + B(1T ) − B(0) dt. (7) ocean water is exchanged with the deep ocean and wT pa- rameterises weakening of the overturning circulation that is We do not explicitly model factors such as the thickness of expected to occur with future climate change (Collins et al., ocean stratification layers, spatial variation of stratification, 2013). nutrient limitations to NPP, or changes in ocean circulation The biological pump refers to the sinking of biomass and due to wind forcing, freshwater forcing, or sea-ice processes organic carbon produced in the upper ocean to deeper ocean (Bernardello et al., 2014). layers (Volk and Hoffert, 1985). In the models on which the IPCC reports are based, a weakening of the biological pump 2.3 Atmosphere is predicted under climate change, mostly due to a decrease in primary production, in turn due to increases in thermal We define cs to be the total carbon in our “system”, com- stratification of ocean waters (Bopp et al., 2013). We rep- prised of carbon stocks in the ocean mixed layer, atmosphere, resent this climate-induced weakening in a single approxi- and terrestrial biosphere, that is mately linear factor, so that the rate of carbon transported out + + = of the upper-ocean mixed layer by the biological pump to ca ct cm cs. (8) lower deep sea layers is given by The only processes that affect the total carbon are human B(1T ) = B0(1 − BT 1T ). emissions of fossil carbon into the atmosphere, e(t), and ex- port of carbon into the deep ocean by the solubility and bio- As on land, we assume a pre-industrial equilibrium at which logical pumps, giving the biological pump was balanced by transport of carbon

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 511

(a) (b) 5 ) ) 1 1 - - 4 r r 4 y y

C C 3 g g P P 2 ( ( 2 Land stock

Ocean stock 0 1 changes changes Historical 0 - 2 1800 1850 1900 1950 2000 2050 2100 1800 1850 1900 1950 2000 2050 2100 RCP2.6 Year Year RCP4.5 (c) (d) 3.5

) 20

1 3.0 -

r RCP6

y 2.5 ) 15 K C ( g 2.0 RCP8.5 P ( 10 1.5 1.0 change 5 Temperature

Atmosphere stock 0.5 changes 0 0.0 1800 1850 1900 1950 2000 2050 2100 1800 1850 1900 1950 2000 2050 2100 Year Year

Figure 2. Model output under forcing from different RCP scenarios: (a) ocean carbon stock change, (b) land carbon stock changes, (c) atmospheric carbon stock change, and (d) global mean surface temperature change. Historical changes in carbon stocks are from Le Quéré et al. (2016) and historical temperature anomalies are from NOAA(2018). The historical temperature dataset of NOAA(2018), which is relative to the period 1901–2000, has been offset to match the model’s average temperature anomaly over the same period.

inition, the initial value of 1T is 0). We choose to parame- dcs terise each process with the best available knowledge about = e(t) − w0(1 − wT 1T )(cm − cm0) dt that process, rather than try to force the model to fit histor- − (B(1T ) − B(0)), (9) ical data. This is in line with our stated model purposes of understanding and learning, rather than prediction. Param- + + in which the initial value of cs is ca0 ct0 cm0. To ob- eters for the response of climate to carbon dioxide (λ, τ) tain the dynamics of atmosphere carbon stocks, we therefore and two parameters of the response of the ocean to chang- solve the differential Eq. (9) and then use the carbon balance ing temperature (BT and wT) were set based on the output of Eq. (8) to find ca. atmosphere–ocean global circulation models. For the climate Increasing atmospheric carbon dioxide levels ca cause a sensitivity λ, transient climate response was used. All other change in global mean surface temperature, 1T , compared parameters are based on historical observations of the global to its pre-industrial level. To model the response of 1T , carbon cycle (Table1). we follow the formulation of Kellie-Smith and Cox(2011), Unless otherwise noted, we perform emissions-based which includes a logarithmic response as per the Arrhenius model runs using harmonised historical data and future RCP law and a delay of timescale τ. Physically, this time delay is scenarios on fossil fuel emissions [e(t)] and land use emis- primarily due to the heat capacity of the ocean. sions [LUC(t)] (Meinshausen et al., 2011b). While the focus d1T 1 λ ca of our study is on future climate change, from the present = log − 1T (10) day until 2100, we begin simulations in 1750 to compare dt τ log2 ca0 our model against historical observations. Time series of the The climate sensitivity λ specifies the increase in temperature model output are displayed in Fig.2. Model solutions were in response to a doubling of atmospheric carbon dioxide lev- approximated in continuous time. els. The climate sensitivity accounts for energy balance feedbacks such as from clouds and albedo. We use the transient climate sensitivity (Collins et al., 2013) as this specifies the 4 Feedback analysis response of the climate system over an approximately 100- year timescale (see Sect.3). Our climate–carbon cycle model is sufficiently simple that the strengths of its feedbacks can be estimated analytically. 3 Model parameterisation and validation Such computations are useful since the resulting symbolic expressions can be used to identify how parameters of in- Our climate–carbon cycle model has 12 parameters, four terest affect feedback strengths and model dynamics. In this state variables, and three nontrivial initial conditions (by def- section we introduce definitions of feedback strengths, cal- www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 512 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

Table 1. Model parameters.

Name Symbol Value Reference/notes

Pre-industrial atmospheric carbon ca0 589 PgC Ciais et al.(2013) Pre-industrial soil and vegetation ct0 1875 PgC 1325 PgC of soil organic carbon in top metre of soil (Köchy et al., 2015) carbon plus mid range of vegetation carbon estimate by the Ciais et al.(2013). Pre-industrial ocean mixed layer cm0 900 PgC Ciais et al.(2013) carbon Climate sensitivity (TCR) λ 1.8 K Multi-model mean transient climate response (Flato et al., 2013) Climate lag τ 4 years Calculations on ocean heat uptake, the primary cause of climate lag, indicate a response time (e-folding time) of 4 years for timescales up to centuries, before deep ocean heat uptake dominates at millennial timescales (Gregory et al., 2015). This result is consistent with simulations that indicate that maximum warming after a CO2 pulse is reached after only a decade (Ricke and Caldeira, 2014) and with results from impulse response model experiments (Joos et al., 2013). Atmosphere–ocean mixed layer D 1 yr−1 Timescale of approximately 1 year, although highly spatially dependent CO2 equilibration rate (Jones et al., 2014). Revelle (buffer) factor r 12.5 Williams et al.(2016) −1 Solubility temperature effect DT 4.23 % K Takahashi et al.(1993); Ciais et al.(2013, p. 498) −1 Pre-industrial biological pump B0 13 PgCyr Ciais et al.(2013) −1 Temperature dependence of bio- BT 3.2 % K 12 % decrease (Bopp et al., 2013, Fig. 9b) after approximately 3.7 K logical pump climate change (Collins et al., 2013) −1 Solubility pump rate w0 0.1 yr DIC ﬂux rate from ocean mixed layer divided by DIC stock in mixed layer (Ciais et al., 2013) −1 Weakening of overturning circula- wT 10 % K Approximate ﬁt to values reported by Collins et al.(2013, p. 1095) tion with climate change Terrestrial respiration temperature QR 1.72 Raich et al.(2002); Xu and Shang(2016). Based on soil respiration, dependence which contributes the majority of terrestrial ecosystem respiration. −1 Pre-industrial NPP NPP0 55 PgCyr Wieder et al.(2015); Sitch et al.(2015) −1 Fertilisation effect KC 0.3 Estimated by substituting recent NPP ≈ 60 PgCyr (Wieder et al., 2015; Sitch et al., 2015) and recent terrestrial carbon stocks, ct ≈ ct0 + 240 (Ciais et al., 2013), into Eq. (1). Alexandrov et al.(2003) found that values between 0.3 and 0.4 are compatible with results from a process-based global NPP model.

on culate climate–carbon cycle feedbacks analytically and nu- Out of the total atmospheric carbon change 1ca , the carbon merically, and estimate feedback non-linearities. cycle feedback contributes (Hansen et al., 1984)

feedback on off 1ca = 1ca − 1ca . (12) Gain is the change in a feedback to atmospheric carbon con- 4.1 Deﬁnitions tent caused by changes in atmospheric carbon content:

There are multiple measures of carbon cycle feedbacks cur- feedback 1ca rently in use. We here review three of the most common mea- g = . (13) 1con sures. a Consider an emission of E Pg C over some time period to Gain and feedback factor are related by the atmosphere. In the absence of carbon cycle feedbacks, the off ≡ 1 atmospheric carbon content would increase by 1ca E. F = . (14) With a feedback switched on, the atmospheric carbon con- 1 − g tent would actually change by 1con. The feedback factor is a An alternative formalism, introduced by Friedlingstein (Zickfeld et al., 2011) et al.(2006), allows feedbacks to be characterised from carbon cycle model output. Climate models are not required, ex- 1con cept as a forcing to the carbon cycle model. The formalism F = a . (11) off relates the changes in terrestrial and marine carbon stocks to 1ca

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 513 changes in global mean temperature and atmospheric carbon we use changes in the equilibrium state of the model to ap- dioxide as follows: proximate model responses over long timescales. We analytically calculate the gains associated with each 1ct = βL1ca + γL1T (15) of the feedback loops in Fig.1 as follows. We calculate the 1cM = βO1ca + γO1T . (16) sensitivity (mathematically, partial derivative) of the equilibrium value of each quantity in the feedback loop with respect Here the βL and βO feedback parameters are the land and to the preceding quantity in the loop. We form the product of ocean, respectively, carbon sensitivities to atmospheric car- the derivatives (as per the chain rule of differentiation) to es- bon dioxide changes 1ca. Likewise, γL and γO are the land timate the gain of that feedback loop. For example, to calcu- and ocean, respectively, carbon sensitivities to temperature late the land climate–carbon gain we calculate the sensitivity changes 1T . Note that cM denotes the total marine carbon of equilibrium temperature with respect to changes in atmo- ∗ stock, both mixed layer and deep ocean. The differences 1ca, spheric carbon content (∂T /∂ca), multiplied by the sensitiv- etc., are usually calculated over the duration of a simulation. ity of equilibrium terrestrial carbon with respect to changes ∗ To isolate the β and γ feedback parameters, simulations are in temperature (∂ct /∂T ), multiplied by the sensitivity of conducted with biogeochemical coupling only and with ra- equilibrium atmospheric carbon with respect to changes in ∗ diative coupling only (Gregory et al., 2009). terrestrial carbon (∂ca /∂ct). In both the formalisms introduced thus far, the feedback Land climate–carbon equilibrium gain. measures are calculated by examining the changes in car- ∗ ∗ ∗ bon stocks at the end point of model simulations. In con- ∗ ∂T ∂ct ∂ca gTL ≡ trast, Boer and Arora(2009) estimate sensitivities 0 and B ∂ca ∂T ∂ct of the instantaneous carbon ﬂuxes from atmosphere to land Land concentration–carbon equilibrium gain. and ocean: ∗ ∗ ∗ ∂ct ∂ca gL ≡ dc ∂ca ∂ct t = B 1c + 0 1T (17) dt L a L Ocean climate–carbon equilibrium gain. dcM ∂T ∗ ∂c ∂c∗ = BO1ca + 0O1T . (18) ∗ ≡ M a dt gTO ∂ca ∂T ∂cM These feedback parameters B and 0 are usually computed Ocean concentration–carbon equilibrium gain. for all time points during a simulation, again using biogeo- ∂c ∂c∗ ∗ ≡ M a chemically coupled and radiatively coupled simulations. gO ∂ca ∂cM The two sets of parameters (B, 0) and (β, γ ) are related by The subscript T denotes that the feedback involves temperature. Asterisks (∗) denote equilibrium quantities. From these Z gains, the feedback factors F ∗ , F ∗, F ∗ , and F ∗ can be cal- = TL L TO O β1ca B1cadt (19) culated using Eq. (14). We label these gain and feedback fac- Z tors g∗ and F ∗, respectively, to denote they are based on an γ 1T = 01T dt. (20) equilibrium approximation, not directly from transient simulations as estimated by Zickfeld et al.(2011). Accordingly, Boer and Arora(2013) refer to B and 0 as di- The derivatives of c∗ are trivial to calculate: by carbon bal- ∗ ∗ a rect feedback parameters and to β and γ as time-integrated ance, ∂ca = ∂ca = −1. To calculate the derivatives of c∗, we ∂ct ∂cM T feedback parameters. dct set 0 = , solve for ct, and calculate the necessary deriva- dt ∗ tives. A similar procedure provides ∂T . 4.2 Analytical feedback strengths based on equilibrium ∂ca The remaining derivatives are ∂cM and ∂cM . Carbon sunk changes ∂T ∂ca into the deep ocean is substantial and cannot be neglected. Analytical approximations to the strengths of carbon cycle Deep ocean carbon storage equilibrates on timescales of mil- feedbacks in our model require choosing a timescale on lennia or more, however, far longer than the timescales of which the feedbacks will be calculated. Numerically esti- interest in this model (we therefore write derivatives of cM ∗ mated feedback factors (Eq. 11) and time-integrated feed- rather than cM). We therefore cannot use the same equilib- back parameters (Eqs. 15–16) are conventionally calculated rium approach as for the other variables. Instead, we derive using carbon stock changes over 100 years or more. Re- approximations to Eq. (7) as follows. First, we observe that sponses on the longest timescales of our model are therefore in the SRES A2 scenario used below, both cm(t) and 1T (t) most relevant if our analytical approximations are to approx- can be approximated as linear increases, starting at cm = cm0 imate numerically calculated values. While recognising that and 1T = 0, respectively, over a time interval tlin. We esti- 0 the Earth’s climate system is presently far from equilibrium, mate this time interval by tlin = (cm(tend)−cm0)/cm(tend) us- www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 514 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

Table 2. Model validation. Historical changes are carbon stocks in 2011 relative to stocks in 1750 (Ciais et al., 2013) and temperatures in 2012 relative to temperatures in 1880 (Hartmann et al., 2013). Predicted future changes are carbon stocks in 2100 compared to 2012 (Collins et al., 2013) and global mean surface temperature (GMST) averaged over 2081–2100 relative to 1986–2005 (Collins et al., 2013), under the range of RCP scenarios.

Ocean carbon changes (Pg C) Land carbon changes (Pg C) GMST change, 1T (K) IPCC AR5 Model result IPCC AR5 Model result IPCC AR5 Model result Historical 155 ± 30 95 −30 ± 45 26 0.85 [0.65 to 1.06] 0.82 RCP2.6 150 [105 to 185] 174 65 [−50 to 195] 67 1.0 [0.3 to 1.7] 0.5 RCP4.5 250 [185 to 400] 243 230 [55 to 450] 135 1.8 [1.1 to 2.6] 1.2 RCP6 295 [265 to 335] 278 200 [−80 to 370] 168 2.2 [1.4 to 3.1] 1.7 RCP8.5 400 [320 to 635] 340 180 [−165 to 500] 207 3.7 [2.6 to 4.8] 2.4

ing the value c and gradient c0 at the end of the simulation ∗ dct 1 m m BL = period. We obtain dt 1T =0 ca − ca0

∗ dcM 1 1 1 0 = 1cM ≈ cm − cm0 + w0 − wT1T cm − cm0 tlin O dt 1T 2 3 ca=ca0

1 ∗ dcM 1 − B0BT 1T tlin. (21) BO = . 2 dt 1T =0 ca − ca0 We use this equation to calculate the derivatives ∂cM and ∂T Here dc /dt and dc /dt denote the atmosphere–land and ∂cM . Evaluating these derivatives will involve the deriva- t M ∂ca atmosphere–ocean fluxes. The subscript 1T = 0 denotes tives ∂cm and ∂cm . Since partial pressures across the air–sea ∂T ∂ca a biogeochemically coupled (and radiatively decoupled) sim- interface equilibrate rapidly on the timescale of the model ulation and ca = ca0 denotes a radiatively coupled (and bio- −1 (D = 1 yr , Table1), we assume that ca ≈ p(cm,1T ), re- geochemically decoupled) simulation. arrange for cm, and then calculate the appropriate derivatives The values of the feedback parameters are strongly sce- from the resulting equation. nario dependent (Arora et al., 2013). To calculate the di- We analytically estimate equilibrium versions of the time- rect feedback parameters, we assume a standard CO2- integrated feedback parameters of Friedlingstein et al.(2006) quadrupling concentration pathway in order to compare our using a similar approach: results with Arora et al.(2013). This scenario has ca(t) = ∗ t = 1 dca = ∂c ca0a , where a 1.01. In this scenario, c dt loga and, ∗ = t a γL dT = ∂T ignoring an initial exponential transient, dt λloga/log2. ∂c∗ For the atmosphere–land carbon flux, the calculation is ∗ = t βL straightforward under the following assumptions. We as- ∂ca sume that NPP0/ct0 loga so that ct tracks its carrying ∗ ∂cM γ = capacity c ≈ K (Eq.2). We also ignore land use change, O ∂T t dct ≈ dK dK | = ∂K dT so that . Then we calculate ca=c and ∗ ∂cM dt dt dt a0 ∂T dt = dK ∂K dca βO . |1T =0 = . ∂c dt ∂ca dt a While the atmosphere–ocean flux could be read off di- Since the ocean component of the model has multiple pro- rectly from the first term of Eq. (6), this form is however cesses that respond to temperature, some analytical forms not particularly useful. As it involves a small difference be- were too complicated for easy visual inspection (Table A1). tween two large quantities, ca and p(cm,1T ), the size of the We derived approximate analytical feedbacks by comparing difference can only be estimated from numerical results and the magnitudes of terms in the numerator and denominator of gives no immediate insight into how it depends on parame- the feedback measures by expanding in power series of DT T ters. Furthermore, we seek to compare our analytical results and ca/ca0. to the results presented by Arora et al.(2013), in which the feedback parameters are presented as functions of ca or 1T 4.3 Analytical feedback strengths based on carbon only (not cm). fluxes We instead derive an approximation based on timescale separation as follows. The characteristic timescale of We estimate the direct feedback parameters as follows: atmosphere–ocean diffusion is much faster than the solubility

∗ dct 1 pump, biological pump, or human emissions into the atmo- 0L = dt 1T sphere (D w0,B0/cm0,loga). Since atmosphere–ocean ca=ca0

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 515

Table 3. Feedback analysis. Gains (g), feedback factors (F ), time-integrated feedback parameters (γ and β), and direct feedback parameters (0 and B) were calculated analytically and numerically. Analytical ocean feedbacks are approximations of the exact forms in Table A1 (see Sect. 4.2). Exact forms were used to calculate numerical values. In this table, p ≡ p(cm,T ). Units of the climate–carbon integrated feedback parameters are Pg C K−1 and concentration–carbon integrated feedback parameters are PgCppm−1. Ranges for analytical results are written in the form (value at start of simulation) to (value at end of simulation). Emissions scenarios are as indicated; land use emissions were assumed to be zero. From the results of simulations using the SRES A2 scenario we use tlin ≈ 100 corresponding to a period between the years 2000 and 2100.

Feedback measure Land climate– Ocean climate– Land conc.– Ocean conc.– carbon feedback carbon feedback carbon feedback carbon feedback λc 1 + K log ca logQ t0 C ca0 R λt B B cmD w c K cmw t Gain, analytical expression lin 0 T + T 0 − t0 C − 0 lin 1T /10 ca log2 2 2r 1T /10 2car 10caQR log2 caQR w w (c − c ) + 0 T m m0 3 Feedback factor (numerical scenario: SRES A2) (> 1 ampliﬁes climate change; < 1 dampens climate change) – estimate from analytical gain 1.81 to 1.18 1.01 to 1.09 0.51 to 0.81 0.89 to 0.84 – from simulation 1.15 1.10 0.80 0.73 – Zickfeld et al.(2011) 1.25 1.22 0.66 0.71 Time-integrated feedback parameter (numerical scenario: SRES A2) (< 0 ampliﬁes climate change; > 0 dampens climate change) t c D w t c logQ − lin − m T 0 lin c w t − t0 R B0BT ct0KC m 0 lin – analytical expression 1T /10 2 2r 10Q tlin ca 2car R −w wt(cm − c ) 0 m0 3 – estimate from analytical form −102 to −86 −3 to −67 2.04 to 0.51 0.26 to 0.42 – from simulation −74 −48 0.84 1.09 – Zickfeld et al.(2011) −129 −32 1.32 0.98 – Friedlingstein et al.(2006) −79 (−20 to −177) −30 (−14 to −67) 1.35 (0.2 to 2.8) 1.13 (0.8 to 1.6)

Direct feedback parameter (numerical scenario: CO2 doubling) (< 0 amplifies climate change; > 0 dampens climate change) ct0λlogQR loga w0cm0DT ct0KC loga w0cm – analytical expression − − − B0BT 101T log2 r ca − ca0 rca – estimate from analytical form Fig. A1a Fig. A1b Fig. A1a Fig. A1b – from simulation Fig. A1a Fig. A1b Fig. A1a Fig. A1b – Arora et al.(2013) see text Non-linearity −0.43 −0.11 0.03 0.03 diffusion is the fastest process, ocean mixed layer carbon back factors from direct numerical simulations of our model. −1 content rapidly gains an equilibrium cm = p (ca,1T ) with To compare the results of our model to previous studies, we −1 respect to atmospheric carbon content, where p (ca,1T ) use the following scenarios. To compare our results with the is the solution to ca = p(cm,1T ). On the timescale of our time-integrated feedback parameters reported by Friedling- model, the atmosphere–ocean flux is therefore controlled by stein et al.(2006) and the feedback factors and gains of the solubility and biological pumps, with diffusion provid- Zickfeld et al.(2011), we employ the SRES A2 emissions ing a rapid coupling between ocean mixed layer and atmo- scenario used in those articles. To compare our results with sphere. An approximation for the atmosphere–ocean flux is the direct feedback parameters of Arora et al.(2013), we use −1 −1 therefore dcMdt ≈ w0(1 − wT 1T )(p (ca,1T ) − cm0) − the same scenario used in that article in which CO2 concen- ∗ −1 B0BT T , which upon substitution into the definitions of BO tration increases 1 % yr until it quadruples (approximately ∗ and 0O gives the forms in Table A1. Taylor series expansions 140 years). For each scenario, we perform four simulations: and L’Hôpital’s rule were then used to derive the approximate forms in Table3. 1. Fully coupled simulation.

off 2. Completely uncoupled simulation, giving ca (t) = 4.4 Numerical estimation of feedback strengths R t ca0 + E(t)dt for the emissions-driven scenario and the In addition to the analytical approximations to carbon cycle speciﬁed concentration pathway for the concentration- feedbacks derived from our model, we also estimate feed- driven scenario. www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 516 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

3. Biogeochemically coupled simulation. We switch off port (Table2). Our model estimates around 55 PgC more feedbacks involving temperature response to atmo- historical land carbon uptake than the IPCC multi-model spheric CO2 by setting λ = 0. Since our model contains mean, possibly due to our simplification to a single land car- no radiative forcing other than changes in CO2, bon pool. Because it omits radiative forcing due to green- temperature 1T = 0 in this simulation. From this sim- house gases other than CO2, our model consistently underes- ulation we estimate the carbon–concentration feedback timates future temperature changes, although in all except the on off off factors via land FL = 1ca /1ca = 1 − 1ct/1ca RCP8.5 scenario the projections are within the IPCC model on off off and ocean FO = 1ca /1ca = 1 − 1cM/1ca , time- range. The purpose of our model is not to precisely predict integrated feedback parameters βL = 1ct/1ca and future climate change, but to serve as an approximate, mech- βO = 1cM/1ca, and direct feedback parameters anistically based emulator of the carbon cycle component = dct − = dcM − BL(t) dt /(ca ca0) and BO(t) dt /(ca ca0). of Earth system models (see Sect.2). If we choose parameters to fit historical observations rather than based on the 4. Radiatively coupled simulation. We switch off feed- best available knowledge about each process (see Sect.3), backs involving the carbon cycle, by setting KC = 0 then our results remain mostly within IPCC model range al- and changing the ca in Eq. (6) to ca0. From this sim- though ocean and land uptake are consistently above and be- ulation we estimate the carbon–climate feedback fac- low the IPCC multi-model mean, respectively (Table A2a). off off tors FTL = 1 − 1ct/1ca and FTO = 1 − 1cM/1ca , We conclude that the model emulates historical observations time-integrated feedback parameters γL = 1ct/1T and and future projections of Earth system models with sufficient γO = 1cM/1T following Arora et al.(2013), and di- accuracy for this purpose. = dct = rect feedback parameters 0L(t) dt /1T and 0O(t) dcM /1T . dt 5.2 Feedback analysis All feedback measures calculated directly from our stylised 4.5 Feedback non-linearity model simulations, as well as most of our analytically es- Zickfeld et al.(2011) found, in emissions-driven scenarios, timated feedback measures, are within a factor of 2 of that the fully coupled simulation sunk more terrestrial and the mean output from Earth system models reported by marine carbon than the sum of the biogeochemically and Friedlingstein et al.(2006) and Zickfeld et al.(2011) (Ta- radiatively coupled scenarios. They refer to this difference ble3; compare also Fig. A1 with Figs. 4–5 of Arora et al. as the non-linearity of the feedback, with the land sink con- (2013) for direct feedback parameters). This is a remarkably tributing 80 % of the non-linearity and the ocean sink 20 %. good agreement considering the highly stylised nature of our Our analytical expressions for the feedbacks can be used to model. Furthermore, all of the numerically time-integrated obtain an alternative measure of feedback non-linearity. feedback parameters from our stylised model are within the We evaluate the non-linearity in the ocean and land multi-model range reported by Friedlingstein et al.(2006). ∗ − climate–carbon feedbacks using FTO(ca,cm,ct,1T ) The agreement observed here serves as additional validation ∗ ∗ − FTO(ca0,cm0,ct0,1T ) and FTL(ca,cm,ct,1T ) of our model as well as validation of the approximations used ∗ FTL(ca0,cm0,ct0,1T ), respectively, where the to calculate analytical feedback factors. ∗ F (ca0,cm0,ct0,1T ) are analytical approximations of An exception to the close agreement is the ocean feedback factors from a radiatively coupled simulation (all concentration–carbon feedback, with the analytically esti- carbon stocks are fixed at pre-industrial levels). We evaluate mated feedback factor and time-integrated feedback parame- the non-linearities in the ocean and land concentration– ter indicating a weaker negative feedback than the numerical ∗ − ∗ carbon feedbacks using FO(ca,cm,ct,1T ) FO(ca,cm,ct,0) estimates from our stylised model or Earth system models. ∗ − ∗ and FL (ca,cm,ct,1T ) FL (ca,cm,ct,0), respectively, where This mismatch is primarily due to two approximations: one ∗ the F (ca,cm,ct,0) are analytical approximations of feed- in the numerical simulation and one in the analytical approx- back factors from a biogeochemically coupled simulation imation. The numerical approximation is that disconnecting (temperature is fixed at its pre-industrial level). These ex- climate feedbacks in the biogeochemically coupled simula- pressions indicate the effect of temperature and atmospheric tion leaves less carbon available to be sunk into the ocean, carbon on the concentration–carbon and climate–carbon placing the ocean feedback at a different point in the highly feedbacks, respectively, We used the SRES A2 scenario. non-linear (as parameterised by the Revelle factor) ocean carbon uptake dynamics. The analytical approximation is that Eq. (21) underestimates carbon sunk into the deep ocean. 5 Results and discussion We used parameters (Table1) based on the best available 5.1 Model evaluation data about each process (see Sect.3). With a set of parameters based instead on fit to historical changes (Table A2), the Most predictions of our model are within the range of model numerically estimated feedbacks became slightly stronger: predictions produced for the IPCC’s Fifth Assessment Re- that is, the already positive climate–carbon feedbacks be-

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 517 came more positive and the already negative concentration– the ocean component of the model, CO2 solubility, the bio- carbon feedbacks more negative. The numerical feedback es- logical pump, and the solubility pump are all temperature de- timates retained, however, good agreement with analytical pendent and therefore contribute to the ocean climate–carbon estimates as well as with previous numerical estimates by feedback. Remarkably, all three processes contribute tem- Friedlingstein et al.(2006) and Zickfeld et al.(2011). One perature dependences of a similar magnitude; we therefore exception was the ocean concentration–carbon feedback, for list all three in the approximate analytical gain and time- which the analytical estimate remained outside Friedling- integrated feedback parameter in Table3. The three terms stein et al.’s range as noted above, but the direct numerical represent temperature dependence of the biological pump, estimate moved to also be outside their range. We conclude CO2 solubility, and the solubility pump. that our results are relatively insensitive to parameter values, though mechanistically based parameter values perform 5.3 Feedback non-linearity slightly better than fitted parameter values. Focusing on the analytical expressions, we observe that As shown in Sect. 4.5, our analytical feedback expressions the approximate analytical expressions for the three dif- enable a new way of estimating feedback non-linearities that ferent feedback measures all have similar dependences on is not possible from direct numerical simulation. Since the state variables and parameters. All measures of the land sum of the four non-linearities is negative (Table3), we con- climate–carbon feedback have dependence of the form clude that summing feedbacks found by individual decou- 1T /10 pled simulations will overestimate the atmospheric carbon ct0 logQR/Q . The ocean climate–carbon feedbacks all R levels, that is, underestimate land and ocean sinks. This result have terms of the form B0BT and w0DT cm/r. The land matches the findings of Zickfeld et al.(2011) and Matthews concentration–carbon feedback has the form ct0KC/ca and the ocean concentration–carbon feedbacks have the form (2007). Terrestrial feedbacks contributed 83 % of the total non-linearity in our model, compared to 80 % reported w0cm/rca. We conclude that for each type of carbon cycle feedback, all three feedback formalisms detect similar fea- by Zickfeld et al.(2011). Furthermore, we can distinguish tures of the climate–carbon system. the non-linearities in the climate–carbon and concentration– The analytical expressions provide rapid insight into how carbon feedbacks. We found that the non-linearity in the ter- feedback strengths depend on state variable and parameter restrial carbon–climate feedback was almost 4 times larger values that could otherwise only be studied numerically or than any other (Table3). By inspecting the analytical deriva- by qualitative reasoning. The analytical forms show that in- tion of the gains we conclude that this dominance is likely creasing Revelle factor r, as is likely to occur in an increas- due to a combination of three reasons: first, due to the sensi- = ingly acidic ocean (Sabine et al., 2004), will decrease the tivity of temperature to carbon dioxide, ∂T /∂ca λ/ca log2, strengths of ocean climate–carbon and concentration–carbon the carbon–climate feedbacks are much more sensitive to ca than the concentration–carbon feedbacks are to 1T . Sec- feedbacks. Weakening overturning circulation, via w0, would also decrease the strength of the ocean carbon cycle feed- ond, the non-linearity in the land climate–carbon feedback is larger than the ocean climate–carbon feedback because backs. On land, the parameters QR and KC control the terrestrial carbon cycle feedbacks. its feedback factor is larger and therefore more sensitive to We can compare likely trends in feedback strengths changes in gain (see Eq. 13). Third, the century timescale from the analytical expressions for the direct feedback pa- of the simulation prevented ocean carbon dynamics, which rameters. According to our model, the destabilising ocean generally take place on longer timescales, from being exhib- climate–carbon feedback is almost constant, while the ocean ited. We conclude that care must be taken when summing results for feedbacks from decoupled simulations, especially concentration–carbon feedback weakens with cm (since 1−r for simulations involving land feedbacks. cm/ca ∼ cm ). Similarly, according to our model the destabilising land climate–carbon feedback will weaken less than the stabilising concentration–carbon feedback (under CO2 6 Conclusions ∼ −1T /10 doubling, QR weakens by 9 % at the new temper- Earth system models span a wide range of complexity. ature equilibrium while ∼ 1/ca weakens by 50 %). This difference between the land climate–carbon and concentration– Here, we constructed a highly stylised, globally aggregated carbon feedbacks stems from the differing curvatures of climate–carbon cycle model. Despite the model’s simplicity – just four state variables – the model emulated globally K(ca,1T ) as a function of 1T (close to linear) and ca (con- cave). We conclude that under continued carbon emissions, aggregated historical trends and future projections of Earth according to our model, both land and ocean feedbacks will system models. The model’s simple form allowed climate– overall become more positive. carbon cycle feedbacks to be estimated analytically, provid- Where multiple processes contribute in parallel to a feed- ing mechanistic insight into these processes. For example, back, inspection of analytical forms can indicate the relative we showed that carbon–climate feedbacks are less sensitive contributions of the different processes to the feedback. In than carbon–concentration feedbacks; on land, this is due to the shape of K(ca,1T ). The simple but accurate climate– www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 518 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing carbon cycle model could be a starting point for model-based Second, the model could comprise a workbench for the investigations of Earth system processes that are too poorly systemic understanding of planetary boundary interactions understood to be incorporated in more comprehensive mod- and hence generate crucial insights into the structure of the els. safe operating space for humanity delineated by the plane- Stylised models such as ours have significant value in pol- tary boundaries (Rockström et al., 2009; Steffen et al., 2015). icy contexts. When confronted with difficult policy decisions Such extensions should focus on linking the core abiotic involving long time periods and significant uncertainty, col- and biotic dimensions of the planetary boundary framework. laborative work with scientists allows policymakers to iden- The present lack of well-developed model representations tify and clarify the impacts of various policy actions. In this of the dynamics and ecosystem structure of biosphere di- context, the utility of a model is to increase stakeholders’ un- versity, heterogeneity, and resilience, despite ongoing efforts derstanding of a system and its dynamics under various con- in this direction (Purves et al., 2013; Bartlett et al., 2016; ditions (Voinov and Bousquet, 2010; Anderies, 2005). This Sakschewski et al., 2016), emphasises the need for a more is in stark contrast to the use of more comprehensive models conceptual understanding of biosphere integrity, its vulner- to predict impacts of policies in which mechanisms under- ability to anthropogenic perturbation, and its role for Earth lying dynamics and trade-offs are not transparent and quick system resilience. explorations with stakeholders are not practical. The utility of a stylised model is in facilitating a learning process rather than in “accurately” predicting outcomes. Data availability. Input parameters are listed in Table 2 and time We foresee at least two strands of valuable future research series inputs are from publicly available data as listed in Sect. 3. based on the climate–carbon cycle model developed in this paper. First, our climate–carbon cycle model could be ex- tended by including further processes relevant on different timescales of interest for Earth system analysis. This would enable a more in-depth analytical analysis of the feedback strengths and gains relating to other aspects of Earth system dynamics, such as the Earth’s energy balance, carbon storage in the tropics compared to extratropics, albedo changes, the cryosphere, nutrient cycles, and even societal feedbacks. The task of characterising the Anthropocene as an epoch could thus move beyond qualitative comparison of human-impact trends to an improved characterisation of the feedbacks that maintain different Earth system “regimes”. The effects on feedback strengths of different functional forms, such as the fertilisation effect KC, and how to constrain these functional forms from data could also be investigated and could yield insight into the continued divergence of Earth system model projections.

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 519

Appendix A

(a) (b) Temperature, ΔT (K)

) 0.06 1

0 1 2 3 4 -

0 r

y 0.05

m ) - 2 0.04 p 1 - p r /

y 0.03 C

- 4 g K P /

( 0.02 C B g - 6

P 0.01 (

Γ - 8 400 600 800 1000

- 10 Atmospheric CO2, ca (ppm)

ΓL (numerical) ΓO (numerical) BL (numerical) BO (numerical)

ΓL (analytical) ΓO (analytical) BL (analytical) BO (analytical)

Figure A1. Direct feedback parameters, (a) climate–carbon feedbacks, and (b) concentration– carbon feedbacks.

Table A1. Exact forms for ocean feedbacks.

Feedback measure Ocean climate–carbon feedback Ocean concentration–carbon feedback h Gain λ B0BT tlin + tlin w w (c − c ) − cm 1 + w t 1 − wT1T ca log2 2 3 0 t m m0 car 0 lin 2 3 i + cmDT 1 + w t 1 − wT1T r(1−DT 1T ) 0 lin 2 3 Time-integrated feedback − cmDT 1 + w t 1 − wT1T cm 1 + w t 1 − wT1T r(1−DT 1T ) 0 lin 2 3 car 0 lin 2 3

− B0BT tlin − tlin − parameter 2 3 w0wt(cm cm0) 1 ! r 1 ca − w0cm0 c 1 w0(1−wT1T )cm0 (1−DT 1T ) r −1 a0 Direct feedback parameter − B B 1T 0 T ca−ca0

www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 520 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

Table A2. Testing parameters ﬁtted to historical data. The following changes to parameter values were made to those listed in Table1: KC = −1 0.25, QR = 2.45, λ = 1.91 K, w0 = 0.185 yr . (a) Historical and projected changes of carbon stocks. See Table2 for further information on how the ﬁgures were calculated and sources for model comparison. (b) Feedback analysis. See Table3 for further information. Analytical forms are omitted here.

(a) Ocean carbon changes (Pg C) Land carbon changes (Pg C) GMST change, 1T (K) IPCC AR5 Model result IPCC AR5 Model result IPCC AR5 Model result Historical 155 ± 30 155 −30 ± 45 −30 0.85 [0.65 to 1.06] 0.85 RCP2.6 150 [105 to 185] 303 65 [−50 to 195] 2 1.0 [0.3 to 1.7] 0.3 RCP4.5 250 [185 to 400] 428 230 [55 to 450] 11 1.8 [1.1 to 2.6] 1.2 RCP6 295 [265 to 335] 484 200 [−80 to 370] 13 2.2 [1.4 to 3.1] 1.7 RCP8.5 400 [320 to 635] 591 180 [−165 to 500] −7 3.7 [2.6 to 4.8] 2.5 (b) Feedback measure Land climate– Ocean climate– Land conc.– Ocean conc.– carbon feedback carbon feedback carbon feedback carbon feedback Feedback factor (numerical scenario: SRES A2) – estimate from analytical gain 4.67 to 1.30 1.01 to 1.16 0.56 to 0.85 0.89 to 0.75 – from simulation 1.27 1.14 0.84 0.60 – Zickfeld et al.(2011) 1.25 1.22 0.66 0.71 Time-integrated feedback parameter (numerical scenario: SRES A2) – estimate from analytical form −168 to −126 −3 to −100 1.70 to 0.38 0.26 to 0.70 – from simulation −119 −60 1.28 1.92 – Zickfeld et al.(2011) −129 −32 1.32 0.98 – Friedlingstein et al.(2006) −79 (−20 to −177) −30 (−14 to −67) 1.35 (0.2 to 2.8) 1.13 (0.8 to 1.6) Non-linearity −1.14 −0.14 0.04 0.04

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 521

Author contributions. SJL, JMA, SEC, JFD, IF, KR, JR, and WS carbon storage to projected twenty-first-century climate change, designed the research. SJL, JFD, IF, TG, and CB constructed the J. Climate, 27, 2033–2053, 2014. model. SJL analysed the model. All authors wrote the paper. Boer, G. and Arora, V.: Temperature and concentration feedbacks in the carbon cycle, Geophys. Res. Lett., 36, L02704, https://doi.org/10.1029/2008GL036220, 2009. Competing interests. The authors declare that they have no con- Boer, G. and Arora, V.: Feedbacks in emission-driven and flict of interest. concentration-driven global carbon budgets, J. Climate, 26, 3326–3341, 2013. Bopp, L., Resplandy, L., Orr, J. C., Doney, S. C., Dunne, J. P., Special issue statement. This article is part of the special issue Gehlen, M., Halloran, P., Heinze, C., Ilyina, T., Séférian, R., “Social dynamics and planetary boundaries in Earth system mod- Tjiputra, J., and Vichi, M.: Multiple stressors of ocean ecosys- elling”. It is not associated with a conference. tems in the 21st century: projections with CMIP5 models, Biogeosciences, 10, 6225–6245, https://doi.org/10.5194/bg-10- 6225-2013, 2013. Ciais, P., Sabine, C., Bala, G., Bopp, L., Brovkin, V., Canadell, J., Acknowledgements. The research leading to these results has Chhabra, A., DeFries, R., Galloway, J., Heimann, M., Jones, C., received funding from the Stordalen Foundation via the Planetary LeQuéré, C., Myneni, R., Piao, S., and Thornton, P.: Carbon and Boundary Research Network (PB.net), the Earth League’s Earth- Other Biogeochemical Cycles, book section 6, Cambridge Uni- Doc programme, the Leibniz Association (project DOMINOES), versity Press, Cambridge, UK and New York, NY, USA, 465– European Research Council Synergy project Imbalance-P (grant 570, https://doi.org/10.1017/CBO9781107415324.015, 2013. ERC-2013-SyG-610028), project grant 2014-589 from the Swedish Collins, M., Knutti, R., Arblaster, J., Dufresne, J.-L., Fichefet, T., Research Council Formas, and a core grant to the Stockholm Friedlingstein, P., Gao, X., Gutowski, W., Johns, T., Krin- Resilience Centre by Mistra. We thank Malin Ödalen for her ner, G., Shongwe, M., Tebaldi, C., Weaver, A., and Wehner, M.: comments on the paper. Long-term Climate Change: Projections, Commitments and Ir- reversibility, book section 12, Cambridge University Press, Edited by: Axel Kleidon Cambridge, UK and New York, NY, USA, 1029–1136, Reviewed by: Martin Heimann and Chris Jones https://doi.org/10.1017/CBO9781107415324.024, 2013. Flato, G., Marotzke, J., Abiodun, B., Braconnot, P., Chou, S., Collins, W., Cox, P., Driouech, F., Emori, S., Eyring, V., References Forest, C., Gleckler, P., Guilyardi, E., Jakob, C., Kattsov, V., Reason, C., and Rummukainen, M.: Evaluation of Cli- Alexandrov, G., Oikawa, T., and Yamagata, Y.: Climate dependence mate Models, book section 9, Cambridge University of the CO2 fertilization effect on terrestrial net primary produc- Press, Cambridge, UK and New York, NY, USA, 741–866, tion, Tellus B, 55, 669–675, 2003. https://doi.org/10.1017/CBO9781107415324.020, 2013. Anderies, J. M.: Minimal models and agroecological policy at the Friedlingstein, P.: Carbon cycle feedbacks and future cli- regional scale: an application to salinity problems in southeastern mate change, Philos. T. Roy. Soc. A, 373, 20140421, Australia, Reg. Environ. Change, 5, 1–17, 2005. https://doi.org/10.1098/rsta.2014.0421, 2015. Anderies, J. M., Carpenter, S. R., Steffen, W., and Rockström, J.: Friedlingstein, P., Cox, P., Betts, R., Bopp, L., Von Bloh, W., The topology of non-linear global carbon dynamics: from tipping Brovkin, V., Cadule, P., Doney, S., Eby, M., Fung, I., Bala, G., points to planetary boundaries, Environ. Res. Lett., 8, 044048, John, J., Jones, C., Joos, F., Kato, T., Kawamiya, M., Knorr, W., https://doi.org/10.1088/1748-9326/8/4/044048, 2013. Lindsay, K., Matthews, H. D., Raddatz, T., Rayner, P., Reick, Arora, V. K., Boer, G. J., Friedlingstein, P., Eby, M., Jones, C. D., C., Roeckner, E., Schnitzler, K.-G., Schnur, R., Strassmann, K., Christian, J. R., Bonan, G., Bopp, L., Brovkin, V., Cadule, P., Weaver, A. J., Yoshikawa, C., and Zeng, N.: Climate–carbon cy- Bala, G., John, J., Jones, C., Joos, F., Kato, T., Kawamiya, M., cle feedback analysis: results from the C4MIP model intercom- Knorr, W., Lindsay, K., Matthews, H. D., Raddatz, T., Rayner, parison, J. Climate, 19, 3337–3353, 2006. P., Reick, C., Roeckner, E., Schnitzler, K.-G., Schnur, R., Strass- Gasser, T., Ciais, P., Boucher, O., Quilcaille, Y., Tortora, M., Bopp, mann, K., Weaver, A. J., Yoshikawa, C., and Zeng, N.: Carbon– L., and Hauglustaine, D.: The compact Earth system model OS- concentration and carbon–climate feedbacks in CMIP5 Earth CAR_v2.2: description and first results, Geosci. Model Dev., 10, system models, J. Climate, 26, 5289–5314, 2013. 271–319, https://doi.org/10.5194/gmd-10-271-2017, 2017a. Bacastow, R., Keeling, C. D., Woodwell, G., and Pecan, E.: Atmo- Gasser, T., Peters, G. P., Fuglestvedt, J. S., Collins, W. J., Shin- spheric carbon dioxide and radiocarbon in the natural carbon cy- dell, D. T., and Ciais, P.: Accounting for the climate–carbon cle, II. Changes from AD 1700 to 2070 as deduced from a geo- feedback in emission metrics, Earth Syst. Dynam., 8, 235–253, chemical model, Tech. rep., Univ. of California, San Diego, La https://doi.org/10.5194/esd-8-235-2017, 2017b. Jolla; Brookhaven National Lab., Upton, NY, USA, 1973. Goodwin, P., Williams, R. G., Follows, M. J., and Dutkiewicz, S.: Bartlett, L. J., Newbold, T., Purves, D. W., Tittensor, D. P., and Ocean-atmosphere partitioning of anthropogenic carbon dioxide Harfoot, M. B.: Synergistic impacts of habitat loss and frag- on centennial timescales, Global Biogeochem. Cy., 21, GB1014, mentation on model ecosystems, P. Roy. Soc. B, 283, 20161027, https://doi.org/10.1029/2006GB002810, 2007. https://doi.org/10.1098/rspb.2016.1027, 2016. Gregory, J. M., Jones, C., Cadule, P., and Friedlingstein, P.: Quanti- Bernardello, R., Marinov, I., Palter, J. B., Sarmiento, J. L., Gal- fying carbon cycle feedbacks, J. Climate, 22, 5232–5250, 2009. braith, E. D., and Slater, R. D.: Response of the ocean natural www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018 522 S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing

Gregory, J. M., Andrews, T., and Good, P.: The incon- Le Quéré, C., Andrew, R. M., Canadell, J. G., Sitch, S., Kors- stancy of the transient climate response parameter under bakken, J. I., Peters, G. P., Manning, A. C., Boden, T. A., Tans, increasing CO2, Philos. T. Roy. Soc. A, 373, 20140417 P. P., Houghton, R. A., Keeling, R. F., Alin, S., Andrews, O. D., https://doi.org/10.1098/rsta.2014.0417, 2015. Anthoni, P., Barbero, L., Bopp, L., Chevallier, F., Chini, L. P., Hansen, J., Lacis, A., Rind, D., Russell, G., Stone, P., Fung, I., Ciais, P., Currie, K., Delire, C., Doney, S. C., Friedlingstein, P., Ruedy, R., and Lerner, J.: Climate sensitivity: analysis of feed- Gkritzalis, T., Harris, I., Hauck, J., Haverd, V., Hoppema, M., back mechanisms, in: Climate Processes and Climate Sensitivity, Klein Goldewijk, K., Jain, A. K., Kato, E., Körtzinger, A., Land- AGU Geophysical Monograph 29, Maurice Ewing Vol. 5, edited schützer, P., Lefèvre, N., Lenton, A., Lienert, S., Lombardozzi, by: Hansen, J. E. and Takahashi, T., Washington, D.C., 130–163, D., Melton, J. R., Metzl, N., Millero, F., Monteiro, P. M. S., 1984. Munro, D. R., Nabel, J. E. M. S., Nakaoka, S.-I., O’Brien, K., Harman, I. N., Trudinger, C. M., and Raupach, M. R.: SCCM– Olsen, A., Omar, A. M., Ono, T., Pierrot, D., Poulter, B., Röden- the Simple Carbon-Climate Model: Technical Documentation, beck, C., Salisbury, J., Schuster, U., Schwinger, J., Séférian, R., CAWCR Technical Report No. 047, The Centre for Australian Skjelvan, I., Stocker, B. D., Sutton, A. J., Takahashi, T., Tian, Weather and Climate Research, Canberra, 2011. H., Tilbrook, B., van der Laan-Luijkx, I. T., van der Werf, G. Hartmann, D., Klein Tank, A., Rusticucci, M., Alexan- R., Viovy, N., Walker, A. P., Wiltshire, A. J., and Zaehle, S.: der, L., Brönnimann, S., Charabi, Y., Dentener, F., Dlugo- Global Carbon Budget 2016, Earth Syst. Sci. Data, 8, 605–649, kencky, E., Easterling, D., Kaplan, A., Soden, B., Thorne, P., https://doi.org/10.5194/essd-8-605-2016, 2016. Wild, M., and Zhai, P.: Observations: Atmosphere and Matthews, H. D.: Implications of CO2 fertilization for future cli- Surface, book section 2, Cambridge University Press, mate change in a coupled climate–carbon model, Global Change Cambridge, UK and New York, NY, USA, p. 159–254, Biol., 13, 1068–1078, 2007. https://doi.org/10.1017/CBO9781107415324.008, 2013. Meinshausen, M., Raper, S. C. B., and Wigley, T. M. L.: Em- Jones, D. C., Ito, T., Takano, Y., and Hsu, W.-C.: Spatial and sea- ulating coupled atmosphere-ocean and carbon cycle models sonal variability of the air-sea equilibration timescale of carbon with a simpler model, MAGICC6 – Part 1: Model descrip- dioxide, Global Biogeochem. Cy., 28, 1163–1178, 2014. tion and calibration, Atmos. Chem. Phys., 11, 1417–1456, Joos, F., Bruno, M., Fink, R., Siegenthaler, U., Stocker, T. F., https://doi.org/10.5194/acp-11-1417-2011, 2011a. Le Quere, C., and Sarmiento, J. L.: An efﬁcient and accurate Meinshausen, M., Smith, S. J., Calvin, K., Daniel, J. S., representation of complex oceanic and biospheric models of an- Kainuma, M., Lamarque, J., Matsumoto, K., Montzka, S., thropogenic carbon uptake, Tellus B, 48, 397–417, 1996. Raper, S., Riahi, K., Thomson, A., Velders, G. J. M., and van Vu- Joos, F., Roth, R., Fuglestvedt, J. S., Peters, G. P., Enting, I. G., uren, D. P. P.: The RCP greenhouse gas concentrations and their von Bloh, W., Brovkin, V., Burke, E. J., Eby, M., Edwards, N. extensions from 1765 to 2300, Climatic Change, 109, 213–241, R., Friedrich, T., Frölicher, T. L., Halloran, P. R., Holden, P. 2011b. B., Jones, C., Kleinen, T., Mackenzie, F. T., Matsumoto, K., Meinshausen, M., Wigley, T. M. L., and Raper, S. C. B.: Em- Meinshausen, M., Plattner, G.-K., Reisinger, A., Segschneider, ulating atmosphere-ocean and carbon cycle models with a J., Shaffer, G., Steinacher, M., Strassmann, K., Tanaka, K., Tim- simpler model, MAGICC6 – Part 2: Applications, Atmos. mermann, A., and Weaver, A. J.: Carbon dioxide and climate im- Chem. Phys., 11, 1457–1471, https://doi.org/10.5194/acp-11- pulse response functions for the computation of greenhouse gas 1457-2011, 2011c. metrics: a multi-model analysis, Atmos. Chem. Phys., 13, 2793– NOAA: Climate at a Glance: Global Time Series, Tech. rep., NOAA 2825, https://doi.org/10.5194/acp-13-2793-2013, 2013. National Centers for Environmental information, available at: Kamiuto, K.: A simple global carbon-cycle model, Energy, 19, 825– http://www.ncdc.noaa.gov/cag/, last access: 16 May 2018. 829, 1994. Purves, D., Scharlemann, J. P., Harfoot, M., Newbold, T., Titten- Kellie-Smith, O. and Cox, P. M.: Emergent dynamics of the sor, D. P., Hutton, J., and Emmott, S.: Ecosystems: time to model climate–economy system in the Anthropocene, Philos. T. Roy. all life on Earth, Nature, 493, 295–297, 2013. Soc. A, 369, 868–886, 2011. Raich, J. W., Potter, C. S., and Bhagawati, D.: Interannual variabil- Köchy, M., Hiederer, R., and Freibauer, A.: Global distribution of ity in global soil respiration, 1980–1994, Global Change Biol., 8, soil organic carbon – Part 1: Masses and frequency distributions 800–812, 2002. of SOC stocks for the tropics, permafrost regions, wetlands, and Randers, J., Golüke, U., Wenstøp, F., and Wenstøp, S.: A user- the world, SOIL, 1, 351–365, https://doi.org/10.5194/soil-1-351- friendly earth system model of low complexity: the ESCIMO 2015, 2015. system dynamics model of global warming towards_2100, Earth Körner, C.: Carbon limitation in trees, J. Ecol., 91, 4–17, 2003. Syst. Dynam., 7, 831–850, https://doi.org/10.5194/esd-7-831- Kriegler, E., Hall, J. W., Held, H., Dawson, R., and Schellnhu- 2016, 2016. ber, H. J.: Imprecise probability assessment of tipping points in Raupach, M. R.: The exponential eigenmodes of the carbon-climate the climate system, P. Natl. Acad. Sci. USA, 106, 5041–5046, system, and their implications for ratios of responses to forcings, 2009. Earth Syst. Dynam., 4, 31–49, https://doi.org/10.5194/esd-4-31- Lenton, T. M.: Land and ocean carbon cycle feedback effects on 2013, 2013. global warming in a simple Earth system model, Tellus B, 52, Ricke, K. L. and Caldeira, K.: Maximum warming occurs about one 1159–1188, 2000. decade after a carbon dioxide emission, Environ. Res. Lett., 9, Lenton, T. M., Held, H., Kriegler, E., Hall, J. W., Lucht, W., Rahm- 124002, https://doi.org/10.1088/1748-9326/9/12/124002, 2014. storf, S., and Schellnhuber, H. J.: Tipping elements in the Earth’s Rockström, J., Steffen, W., Noone, K., Persson, Å., Chapin, F. S., climate system, P. Natl. Acad. Sci. USA, 105, 1786–1793, 2008. Lambin, E. F., Lenton, T. M., Scheffer, M., Folke, C., Schellnhu-

Earth Syst. Dynam., 9, 507–523, 2018 www.earth-syst-dynam.net/9/507/2018/ S. J. Lade et al.: Analytically tractable climate–carbon cycle feedbacks under 21st century anthropogenic forcing 523

ber, H. J., Nykvist, B., de it, C. A., Hughes, T., van der Leeuw, Takahashi, T., Olafsson, J., Goddard, J. G., Chipman, D. W., and S., Rodhe, H., Sörlin, S., Snyder, P. K., Costanza, R., Svedin, Sutherland, S. C.: Seasonal variation of CO2 and nutrients in the U., Falkenmark, M., Karlberg, L., Corell, R. W., Fabry, V. J., high-latitude surface oceans: a comparative study, Global Bio- Hansen, J., Walker, B., Liverman, D., Richardson, K., Crutzen, geochem. Cy., 7, 843–878, 1993. P., and Foley, J. A.: A safe operating space for humanity, Nature, Voinov, A. and Bousquet, F.: Modelling with stakeholders, Environ. 461, 472–475, 2009. Modell. Softw., 25, 1268–1281, 2010. Sabine, C. L., Feely, R. A., Gruber, N., Key, R. M., Lee, K., Volk, T. and Hoffert, M. I.: Ocean carbon pumps: Analysis of Bullister, J. L., Wanninkhof, R., Wong, C., Wallace, D. W., relative strengths and efﬁciencies in ocean-driven atmospheric Tilbrook, B., Millero, F. J., Peng, T.-H., Kozyr, A., Ono, T., and CO2 changes, in: The Carbon Cycle and Atmospheric CO: Nat- Rios, A. F.: The oceanic sink for anthropogenic CO2, Science, ural Variations Archean to Present, edited by: Sundquist, E. T. 305, 367–371, 2004. and Broecker, W. S., American Geophysical Union, Washington, Sakschewski, B., von Bloh, W., Boit, A., Poorter, L., Peña- D.C., 99–110, 1985. Claros, M., Heinke, J., Joshi, J., and Thonicke, K.: Resilience Wieder, W. R., Cleveland, C. C., Smith, W. K., and Todd-Brown, K.: of Amazon forests emerges from plant trait diversity, Nat. Clim. Future productivity and carbon storage limited by terrestrial nu- Change, 6, 1032–1036, https://doi.org/10.1038/nclimate3109, trient availability, Nat. Geosci., 8, 441–444, 2015. 2016. Williams, R. G., Goodwin, P., Roussenov, V. M., and Bopp, L.: Schellnhuber, H. J., Rahmstorf, S., and Winkelmann, R.: Why the A framework to understand the transient climate re- right climate target was agreed in Paris, Nat. Clim. Change, 6, sponse to emissions, Environ. Res. Lett., 11, 015003, 649–653, 2016. https://doi.org/10.1088/1748-9326/11/1/015003, 2016. Sigman, D. M. and Boyle, E. A.: Glacial/interglacial variations in Xu, M. and Shang, H.: Contribution of soil respiration to the global atmospheric carbon dioxide, Nature, 407, 859–869, 2000. carbon equation, J. Plant Physiol., 203, 16–28, 2016. Sitch, S., Friedlingstein, P., Gruber, N., Jones, S. D., Murray- Zaehle, S., Friedlingstein, P., and Friend, A. D.: Terrestrial nitrogen Tortarolo, G., Ahlström, A., Doney, S. C., Graven, H., Heinze, feedbacks may accelerate future climate change, Geophys. Res. C., Huntingford, C., Levis, S., Levy, P. E., Lomas, M., Poul- Lett., 37, L01401, https://doi.org/10.1029/2009GL041345, 2010. ter, B., Viovy, N., Zaehle, S., Zeng, N., Arneth, A., Bonan, Zickfeld, K., Eby, M., Matthews, H. D., Schmittner, A., and G., Bopp, L., Canadell, J. G., Chevallier, F., Ciais, P., Ellis, Weaver, A. J.: Nonlinearity of carbon cycle feedbacks, J. Cli- R., Gloor, M., Peylin, P., Piao, S. L., Le Quéré, C., Smith, B., mate, 24, 4255–4275, 2011. Zhu, Z., and Myneni, R.: Recent trends and drivers of regional sources and sinks of carbon dioxide, Biogeosciences, 12, 653– 679, https://doi.org/10.5194/bg-12-653-2015, 2015. Steffen, W., Richardson, K., Rockström, J., Cornell, S. E., Fetzer, I., Bennett, E. M., Biggs, R., Carpenter, S. R., de Vries, W., de Wit, C. A., Folke, C., Gerten, D., Heinke, J., Mace, G. M., Persson, L. M., Ramanathan, V., Rey- ers, B., and Sörlin, S.: Planetary boundaries: Guiding human development on a changing planet, Science, 347, 1259855, https://doi.org/10.1126/science.1259855, 2015.

www.earth-syst-dynam.net/9/507/2018/ Earth Syst. Dynam., 9, 507–523, 2018